, 2011) and the task structure required differences in firing rates in the two populations during target achievement. We therefore performed a thinning procedure to equate firing rates in the two populations (Gregoriou et al., 2009; Experimental Procedures).

Despite differences in firing rate being removed, there remained a significant difference in spike-field FG-4592 mw coherence between output cells and indirect cells (Figure S3; p < 0.001, Bonferroni corrected), demonstrating that this effect was not driven by firing rate differences. To further ensure that our results were not affected by firing rate, we separated our analysis by cell and trial type to examine trials in which output cells were required to increase their firing rate to achieve the target and trials in which output cells decreased their firing rate (Figure S3). There was still a significant difference in coherence between output cells that decreased their firing rate relative to indirect cells (p < 0.05, Bonferroni corrected), despite no significant difference BLZ945 in firing rate between these populations. Finally, we also calculated

coherence after removing cells with low signal-to-noise ratio (SNR) from the indirect population and coherence remained higher in output cells than indirect cells, demonstrating that the effect was not due to differences in SNR (Figure S3; p < 0.05, Bonferroni corrected). These coherent interactions were greatly diminished between trials when rats were not actively engaged in the task (Figure 3D). Furthermore, during these periods, the difference in coherence between output and indirect populations

was abolished (Figures 3E and 3F). These results show that the corticostriatal coherence that emerged during learning was highly specific for neurons that are directly relevant to behavioral output, even when they are closely intermingled with other cells, and that these precise interactions are flexible and appear rapidly as needed during task performance. Because we about found that M1 spikes occurred preferentially at the peak of the DS LFP (Figure 2B), we next investigated the phase offset of the spike-field coherence. From the mean phase heat map, we see that there is a consistent negative phase offset in the 6–14 Hz range (Figure 4A). By convention, this suggests that M1 spikes precede the peak of the DS LFP in the 6–14 Hz band. Indeed, the phase at 6–14 Hz was commonly negative, as can be seen in the distribution of phase offsets for every cell and every frequency from 6 to 14 Hz (Figure 4B). When phase offset values are used to estimate a temporal delay between M1 spikes and DS LFP (see Experimental Procedures), we see a clear preference for M1 cells to fire at an offset of −5 to −7 ms relative to the DS LFP, as reflected in the mode of this distribution (Figure 4C; SEM = 0.03 ms).